Preface

The following notes were prepared by me (Oliver Maclaren) and Ruanui Nicholson for ENGSCI 205: Engineering-Centric Machine Learning, based on discussions with Piaras Kelly, Justin Fernandez, Cameron Walker, Andreas Kempa-Liehr, Ivan Koptev, and others.

They assume some basic familiarity with fitting simple statistical machine learning models using e.g. scikit-learn. The focus is on high-level ideas about what we put into these models, as well as on developing practical familiarity working with multi-dimensional arrays called tensors in PyTorch in order to construct basic linear and nonlinear (neural) models for oneself. After initial introductory material, we focus on incorporate engineering or scientific insights via penalty or regularisation terms in linear and nonlinear models.

The plan is roughly

• Week 1: Key ideas
• Week 2: Vectors, matrices, and tensors in PyTorch
• Week 3: Making linear data models yourself
• Week 4: Incorporating scientific knowledge in penalty terms
• Week 5: Making nonlinear data models (neural networks)
• Week 6: Incorporating scientific knowledge in neural networks